Bruhovv·4h agoHow I Made Local Transcription 4.4× Faster on a MacBook AirA Voice Memos workflow with whisper.cpp, pyannote and a GPU setting that made the difference.
Raymond Ng, MH·14h agoServing Qwen3-ASR: llama.cpp vs vLLM on an NVIDIA GB10For a shared Cantonese speech-recognition server, llama.cpp’s 8-bit model made fewer errors than vLLM, answered faster, handled more…
Daniel James·1d agoMy Voice Memos Were a Graveyard Until I Switched to Speech to TextI get my best ideas while driving, which is the worst time to have them. So for years I talked into my phone, felt very smart, and never…A response icon2A response icon2
Sachin Chaurasiya·1d agoVoice Agents Deep Dive Part 0: The Voice AI Landscape and How Machines Hear and SpeakMost developers picture voice AI as a chatbot with a microphone, then hit a wall when latency, audio, and speech recognition all have to…
InSciforcebySciforce·1d agoVoice AI in Insurance: Which Calls to Automate and How to Get Them RightSomeone who has just been rear-ended tends to call their insurer from the side of the road, still shaken, with questions in no particular…
Raymond Ng, MH·1d agoCantonese Speech Recognition on a Windows PC: NVIDIA, an “Unsupported” AMD Card, and the CPUQwen3-ASR ran on an 8 GB Radeon that ROCm doesn’t support, as fast as on our NVIDIA GB10, through llama.cpp and Vulkan. On the same NVIDIA…
Raymond Ng, MH·3d agoQwen3-ASR vs Whisper for Cantonese: Accuracy and Speed on GPU and CPUWe hoped faster-whisper could replace a vLLM-served model for Cantonese speech recognition. On 500 labelled clips, Qwen3-ASR made about…A response icon1A response icon1
Adnan Masood, PhD.·3d agoHow Voice Models Work From Whisper to Conversational AISpeech recognition, voice generation, model architectures, and the open and proprietary systems behind conversational AI.
Raymond Ng, MH·4d agoWhisper large-v3 on GB10: CPU, GPU, and INT8 Compared for CantoneseOur first test found an 18× CPU/GPU gap. Then we tried faster-whisper — and discovered that the language setting deserved attention too.
vinay krishna·5d agoI Tested Free Offline AI Transcription on a Normal LaptopWhisper offline on my CPU: 48s of speech in under 2s with tiny. Base invented a word I never said.A response icon1A response icon1
